BoseungJeong / ASMR-ICCV-2021

This repository is official implementation of "ASMR: Learning Attribute-Based Person Search with Adaptive Semantic Margin Regularizer" (ICCV 2021)

ASMR: Learning Attribute-Based Person Search with Adaptive Semantic Margin Regularizer

This repository is official implementation of "ASMR: Learning Attribute-Based Person Search with Adaptive Semantic Margin Regularizer" (ICCV 2021).

Overall architecture

Requirements

• Python == 3.7.7
• Pytorch == 1.6.0

Installation

git clone https://github.com/BoseungJeong/ASMR.git
cd ASMR
conda env create -f ASMR.yaml
conda activate ASMR

Download data

Market-1501: Download original Market-1501 dataset and its attribute dataset.

Pre-processing data

After downloading Market-1501, the pre-processing is required to run our model by following:

python prepare_Market.py
python prepare_Market_Attribute.py

GPU information

• NVIDIA Titan XP

Training and Evaluation

After installing requirements and downloading datasets, you can run our model by following:

• Pre-training with attribute recognition

python train_AR.py --AR --train_all --lr 0.01 --batchsize 64 --num_epoch 40

• Train and evaluation on Market-1501

sh Train_and_Test_ASMR_Market.sh

Citation

If you find ASMR useful in your work, you can cite our paper:

@InProceedings{Jeong_2021_ICCV,
author    = {Jeong, Boseung and Park, Jicheol and Kwak, Suha},
title     = {ASMR: Learning Attribute-Based Person Search With Adaptive Semantic Margin Regularizer},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month     = {October},
year      = {2021},
pages     = {12016-12025}
}